Complex and heterogeneous multi-level data are collected in many scientific fields, including genomics and human nutrition.
Such studies, which often focus on comparisons between a pair of experimental groups, often include measures of diverse
natures, including quantitative, semi-quantitative, and/or qualitative variables (i.e., mixed-type data). A major scientific
question of interest focuses on the inference of context-specific relationships, or networks, among experimental variables;
however, existing network inference approaches for mixed-type data typically rely on the use of data transformations or
overly-simplistic model approximations. In this project, we aim at (1) proposing and implementing a core unifying and broadly
applicable multivariate framework for differential network analysis from mixed-type data based on randomized pairwise
likelihoods and copulas; (2) extending this method to address application-specific challenges; and (3) applying this framework
to perform a novel analysis of mixed-type data in three widely different applications spanning major research themes across
multiple INRAE departments (human nutrition, plant genetics, animal genetics). Throughout, our proposed multivariate
mixed-data differential network model serves as a unifying thread, and we will seek to leverage this approach in close
collaboration with domain experts to extract deeper biological knowledge from existing datasets generated across disciplines
within INRAE.